An Agent-Based Parallel Ant Algorithm with an Adaptive Migration Controller

This paper presents an agent-based parallel ant algorithm (APAA) for numerical optimization. By utilizing two cooperating agents to optimize different components of the solution vector, APAA reduces the scale of the problem handled by each agent and thus achieves improvement in algorithm performance. In APAA, each agent has tunable and untunable vectors. Tunable vectors are optimized through a novel continuous ant algorithm with untuable vector fixed. Excellent tunable vectors in one agent are migrated into the other agent as new untunable vectors through a stagnation-based asynchronous migration controller (SAMC), in which the migration strategy is adaptively adjusted according to the stagnation degree in the optimization process of the agent. The proposed APAA is especially suitable for large-scale problems. Experimental studies on a set of benchmark functions show that APAA can obtain better results at a faster speed for functions in high dimensional space.